ABSTRACT

The identifiability problems in competing risk data arise because only the first failure to occur on any subject is observed, this effectively censoring the remaining time or times to failure. One approach to the construction of bivariate survival distributions is to generalize the lack of memory property of the exponential distribution. The bivariate shock model of is due to Marshall and Olkin. Fully nonparametric estimation of a bivariate survivor function when either or both components may be censored is best viewed as an application of the em algorithm. Tests for dependence between two variables when either or both may be subject to censoring have been considered by many authors. If the censoring mechanisms for the two components are independent, or if there is censoring in only one component, an exact permutation distribution of the test statistic under the null hypothesis can be obtained.